Amazon Web Services (AWS) — at AWS re:Invent — introduced two custom computing chips aimed at helping its customers beat the cost of using chips from Intel and Nvidia.
With $45.37 billion in sales in 2020, Amazon Web Services (AWS) is the world’s biggest cloud computing provider and one of the biggest buyers of data center chips, whose computing power AWS rents out to its customers.
Amazon.com’s cloud computing unit on Tuesday released the third generation of its Graviton chip that is designed to compete with central processors from Intel and AMD. The Graviton3 is 25 percent faster than its predecessor.
Dave Brown, vice president of Elastic Compute Cloud at Amazon, told Reuters that the company expects it to provide a better performance per dollar than Intel’s chips.
AWS also said that a new class of chip called Trainium, which is designed to train machine learning computer models and will compete against chips from Nvidia, will soon be available to its customers. AWS expects it to train machine learning models for a cost that is 40 percent lower than Nvidia’s flagship chip.
AWS still works closely with Intel, AMD and Nvidia – for example, it is working with Nvidia to pair its Graviton processors to providing a way for the Android game developer to stream its titles to devices. Brown said AWS wants to keep the computing market competitive by offering an additional chip choice.
Raj Bala, a vice president and analyst at research firm Gartner, said the chip companies should take competition from AWS seriously over the long term.
For now, many cloud computing customers will want to use Intel and Nvidia chips because decades of software have been written to run on them. Only early adopters who can handle the complexity of re-writing their own software are likely to try the new AWS chips, Bala said.
But the same was true when AWS launched a decade and half ago and was used by smaller tech-savvy customers. The company eventually expanded to mainstream companies and is now on pace to become as large as traditional firms like Cisco Systems.
New C7g instances powered by AWS Graviton3 processors deliver up to 25 percent better performance than current generation C6g instances powered by AWS Graviton2 processors.
New Trn1 instances powered by AWS Trainium chips provide the best price performance and the fastest time to train most machine learning models in Amazon EC2. New storage-optimized Im4gn/Is4gen/I4i instances based on AWS-designed AWS Nitro SSDs (solid-state drives) offer the best storage performance for I/O-intensive workloads running on Amazon EC2.
Customers like DirecTV, Discovery, Epic Games, Formula 1, Honeycomb.io, Intuit, Lyft, MercardoLibre, NextRoll, Nielsen, SmugMug, Snap, Splunk, and Sprinklr have seen significant performance gains and reduced costs from running AWS Graviton2-based instances in production since they launched in 2020. The Graviton2 instance portfolio offers 12 different instances that include general purpose, compute optimized, memory optimized, storage optimized, burstable, and accelerated computing instances, so customers have the deepest and broadest choice of cost-effective and power-efficient compute in the cloud.
C7g instances, powered by next generation AWS Graviton3 processors, provide up to 25 percent better performance for compute-intensive workloads compared to current generation C6g instances powered by Graviton2 processors.
AWS Graviton3 processors also deliver up to 2x higher floating point performance for scientific, machine learning, and media encoding workloads, up to 2x faster performance for cryptographic workloads, and up to 3x better performance for machine learning workloads compared to previous generation AWS Graviton2 processors.
AWS Graviton3 processors are also more energy efficient, using up to 60 percent less energy for same performance than comparable EC2 instances. C7g instances are the first in the cloud to feature the latest DDR5 memory, which provides 50 percent higher memory bandwidth versus AWS Graviton2-based instances to improve the performance of memory-intensive applications like scientific computing.
C7g instances also deliver 20 percent higher networking bandwidth compared to AWS Graviton2-based instances. C7g instances support Elastic Fabric Adapter (EFA), which allows applications to communicate directly with network interface cards, providing lower and more consistent latency, to enhance the performance of applications that require parallel processing at scale like HPC and video encoding. C7g instances are available today in preview. To learn more about C7g instances, visit aws.amazon.com/ec2/instance-types/c7g.
Trn1 instances powered by AWS Trainium chips offer the best price performance and the fastest machine learning model training in Amazon EC2, providing up to 40 percent lower cost to train deep learning models compared to the latest P4d instances.
Trn1 instances offer 800 Gbps EFA networking bandwidth (2x higher than the latest EC2 GPU-based instances) and integrate with Amazon FSx for Lustre high performance storage — enabling customers to launch Trn1 instances with EC2 UltraClusters capability.
Trn1 instances are available today in preview. To learn more about Trn1 instances, visit aws.amazon.com/ec2/instance-types/trn1.
SAP HANA is a world’s leading in-memory database that serves as the foundation of the SAP Business Technology Platform. “Over the past decade, SAP HANA has helped customers manage their most mission critical transactional and analytics workloads,” said Irfan Khan, President of HANA Database & Analytics at SAP. “AWS investments and innovations on ARM-based AWS Graviton processors and SAP HANA Cloud are a great match with potential to deliver step-wise operation and performance improvement benefits to our enterprise customers, and to SAP’s cloud analytics and data management solutions powered by SAP HANA Cloud.”
Twitter is what’s happening and what people are talking about right now. “Twitter is working on a multi-year project to leverage the AWS Graviton-based EC2 instances to deliver Twitter timelines. As part of our ongoing engineering to drive further efficiencies, we tested the new Graviton3-based C7g instances,” said Nick Tornow, Head of Platform at Twitter.
“Across a number of benchmarks that we’ve found to be representative of the performance of Twitter workloads, we found Graviton3-based C7g instances deliver 20 percent-80 percent higher performance versus Graviton2-based C6g instances, while also reducing tail latencies by as much as 35 percent. We are excited to utilize Graviton3-based instances in the future to realize significant price performance benefits.”
Formula 1 (F1) racing began in 1950 and is the world’s most prestigious motor racing competition, as well as the world’s most popular annual sporting series. “We had already seen that Graviton2-based C6gn instances provided us the best price performance for some of our CFD workloads. We have now found Graviton3 C7g instances to be 40 percent faster than the Graviton2 C6gn instances for those same simulations,” said Pat Symonds, CTO at Formula 1 Management.
Epic Games is the creator of Fortnite, Unreal, Gears of War, Shadow Complex, and the Infinity Blade series of games.
“Our testing has shown AWS Graviton3-based EC2 instances are suitable for even the most demanding, latency-sensitive workloads while providing significant price performance benefits and expanding what is possible within Fortnite and any Unreal Engine created experience,” said Mark Imbriaco, Senior Director of Engineering at Epic Games.
Honeycomb develops an observability platform that enables engineering teams to visualize, analyze, and improve cloud application quality and performance.
“We were able to run 30 percent fewer instances of C7g than C6g serving the same workload, and with 30 percent reduced latency. We are looking forward to adopting AWS Graviton3-powered C7g instances in production once they are generally available,” said Liz Fong-Jones, Principal Developer Advocate at honeycomb.io, said.
Anthropic builds AI systems that will have many opportunities to create value commercially and for public benefit.
“We are looking forward to using Trn1 instances powered by AWS Trainium, as their unprecedented ability to scale to tens of thousands of nodes and higher network bandwidth will enable us to iterate faster while keeping our costs under control,” Tom Brown, Co-founder at Anthropic, said.
Splunk is a leading data platform provider and is designed to investigate, monitor, analyze, and act on data at any scale.
“We run C/C++ based workloads for indexing and searching event data. Our workload is CPU bound and benefits from high capacity and low latency SSD storage,” said Brad Murphy, Vice President, Cloud Platform & Infrastructure at Splunk. “When evaluating the new Im4gn/Is4gen instances powered by AWS Graviton2, we observed an up to 50 percent decrease in search runtime compared to I3/I3en instances, which we currently use. This makes Im4gn and Is4gen instances a great choice for running our storage-intensive workloads with significant price performance improvement and lower TCO.”
Sprinklr helps the world’s biggest companies make their customers happier across 30+ digital channels—using the sophisticated AI engine.
“We benchmarked our Java-based search workloads on Amazon EC2 Im4gn/Is4gen instances powered by AWS Graviton2 processors. Smaller Is4gen instances offer similar performance compared to larger I3en instances, presenting an opportunity to meaningfully reduce the TCO,” said Abhay Bansal, Vice President of Engineering at Sprinklr. “We also saw a significant 50 percent reduction in latency for queries when moving our workloads from I3 to Im4gn instances, indicating a significant 40 percent price performance benefit.”
Redis Enterprise powers mission critical apps and services for over 8,000 organizations by enabling software teams to create a high-performance data layer for the real-time world.
“We’re thrilled to see the Amazon EC2 I4i instances using the new low latency AWS Nitro SSDs that deliver better transaction speed than the previous generation instances,” said Yiftach Shoolman, Co-Founder and CTO at Redis. “We expect the faster storage performance and higher networking and processor speeds of the I4i instances will deliver significant improvements at an even more attractive total cost of ownership for our customers who use Redis-on-Flash on I4i instances.”
